The proposed research aims to solve important analytical problems arising in almost every clinical study, in which the endpoint is the time to a certain clinical or biological event (e.g., the time to death, AIDS, HIV-RNA failure, etc.). In particular, this project proposes to address the following specific aims: 1. To develop model and variable selection techniques based on aggregate, clinically-meaningful prediction error in regression models for binary, continuous, and censored event outcomes; 2. To study methods for assessing the added value of expensive or invasive markers for predicting outcome or diagnosing disease in HIV and other infectious disease studies using the prediction criteria developed and studied in Aim 1; 3. To develop new sensitivity analysis tools for survival, analysis in the presence of informative censoring; and 4. To develop efficient combinations of a class of estimates for survival models for non-proportional hazards semi-parametric models with censored failure time data. Relevance to Public Health Important progress has been made in the last 15 years in HIV research, resulting in an increase in the treatment options for HIV-infected individuals. The increasingly detailed biologic measurements in HIV, such as genetic mutations in the virus or information in a patient's genome, now provide the opportunity not only for estimating population effects, but also for developing models for predicting individual patient outcomes, including potential toxic side effects of therapy, biologic response, or the time to clinical or biologic failure. The results of this research are expected to assist clinicians in providing better care and case management for patients. [unreadable] [unreadable] [unreadable]